Fig. 4
From: Sparse self-attention aggregation networks for neural sequence slice interpolation
Illustration of how the three terms in the style balanced loss evolves during training. Here, sign0,sign1,sign2 denote \(sign\left (\mathcal {L}_{style_{1'0}}-\mathcal {L}_{style_{10}}\right)\mathcal {L}_{style_{1'0}}, sign\left (\mathcal {L}_{style_{1'1}}-\mathcal {L}_{style_{11}}\right)\mathcal {L}_{style_{1'1}}\) and \(sign\left (\mathcal {L}_{style_{1'2}}-\mathcal {L}_{style_{12}}\right)\mathcal {L}_{style_{1'2}}\), respectively